-
Notifications
You must be signed in to change notification settings - Fork 40
/
k_fold.py
130 lines (106 loc) · 5.6 KB
/
k_fold.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
import os
import re
from copy import copy
from typing import Tuple, Union, Dict
import pandas as pd
from sklearn import model_selection
from torcheeg.datasets.module.base_dataset import BaseDataset
class KFold:
r'''
A tool class for k-fold cross-validations, to divide the training set and the test set. One of the most commonly used data partitioning methods, where the data set is divided into k subsets, with one subset being retained as the test set and the remaining k-1 being used as training data. In most of the literature, K is chosen as 5 or 10 according to the size of the data set.
:obj:`KFold` devides subsets without grouping. It means that during random sampling, adjacent signal samples may be assigned to the training set and the test set, respectively. When random sampling is not used, some subjects are not included in the training set. If you think these situations shouldn't happen, consider using :obj:`KFoldPerSubjectGroupbyTrial` or :obj:`KFoldGroupbyTrial`.
.. image:: _static/KFold.png
:alt: The schematic diagram of KFold
:align: center
|
.. code-block:: python
cv = KFold(n_splits=5, shuffle=True, split_path='./split')
dataset = DEAPDataset(io_path=f'./deap',
root_path='./data_preprocessed_python',
online_transform=transforms.Compose([
transforms.To2d(),
transforms.ToTensor()
]),
label_transform=transforms.Compose([
transforms.Select(['valence', 'arousal']),
transforms.Binary(5.0),
transforms.BinariesToCategory()
]))
for train_dataset, test_dataset in cv.split(dataset):
train_loader = DataLoader(train_dataset)
test_loader = DataLoader(test_dataset)
...
Args:
n_splits (int): Number of folds. Must be at least 2. (default: :obj:`5`)
shuffle (bool): Whether to shuffle the data before splitting into batches. Note that the samples within each split will not be shuffled. (default: :obj:`False`)
random_state (int, optional): When shuffle is :obj:`True`, :obj:`random_state` affects the ordering of the indices, which controls the randomness of each fold. Otherwise, this parameter has no effect. (default: :obj:`None`)
split_path (str): The path to data partition information. If the path exists, read the existing partition from the path. If the path does not exist, the current division method will be saved for next use. (default: :obj:`./split/k_fold_dataset`)
'''
def __init__(self,
n_splits: int = 5,
shuffle: bool = False,
random_state: Union[None, int] = None,
split_path: str = './split/k_fold_dataset'):
self.n_splits = n_splits
self.shuffle = shuffle
self.random_state = random_state
self.split_path = split_path
self.k_fold = model_selection.KFold(n_splits=n_splits,
shuffle=shuffle,
random_state=random_state)
def split_info_constructor(self, info: pd.DataFrame) -> None:
for fold_id, (train_index,
test_index) in enumerate(self.k_fold.split(info)):
train_info = info.iloc[train_index]
test_info = info.iloc[test_index]
train_info.to_csv(os.path.join(self.split_path,
f'train_fold_{fold_id}.csv'),
index=False)
test_info.to_csv(os.path.join(self.split_path,
f'test_fold_{fold_id}.csv'),
index=False)
@property
def fold_ids(self):
indice_files = list(os.listdir(self.split_path))
def indice_file_to_fold_id(indice_file):
return int(re.findall(r'fold_(\d*).csv', indice_file)[0])
fold_ids = list(set(map(indice_file_to_fold_id, indice_files)))
fold_ids.sort()
return fold_ids
def split(self, dataset: BaseDataset) -> Tuple[BaseDataset, BaseDataset]:
if not os.path.exists(self.split_path):
os.makedirs(self.split_path)
self.split_info_constructor(dataset.info)
fold_ids = self.fold_ids
for fold_id in fold_ids:
train_info = pd.read_csv(
os.path.join(self.split_path, f'train_fold_{fold_id}.csv'))
test_info = pd.read_csv(
os.path.join(self.split_path, f'test_fold_{fold_id}.csv'))
train_dataset = copy(dataset)
train_dataset.info = train_info
test_dataset = copy(dataset)
test_dataset.info = test_info
yield train_dataset, test_dataset
@property
def repr_body(self) -> Dict:
return {
'n_splits': self.n_splits,
'shuffle': self.shuffle,
'random_state': self.random_state,
'split_path': self.split_path
}
def __repr__(self) -> str:
# init info
format_string = self.__class__.__name__ + '('
for i, (k, v) in enumerate(self.repr_body.items()):
# line end
if i:
format_string += ', '
# str param
if isinstance(v, str):
format_string += f"{k}='{v}'"
else:
format_string += f"{k}={v}"
format_string += ')'
return format_string